Moody B&W with DSLR

Started Nov 27, 2008 | Discussions thread
GordonBGood
Veteran MemberPosts: 6,278
Like?
Re: 16bit not
In reply to DSPographer, Dec 3, 2008

DSPographer wrote:

I hope you don't mind my answering as well. I have also been thinking
about a compression method as you describe. One possibility (which
would allow the compressed file to retain the camera identity for
calibration puposes) is to just replace the full linear value with
the nearest quantized linear value. So there are less codes (like if
the lower bits are all zero) but the number of bits in the codes
stays the same. Then, when the lossless jpg step is used to store the
raw data the data will have less stochastic content which should also
reduce the file size. How well this works compared with a NEF style
of replacing the data with a shorter code with less bits would need
to be evaluated but the fact the camera identity would be unchanged
means that the DNG converter would use the correct color calibration
etc.

I don't mind at all, as long as the OP doesn't.

Zeroing the noise filled lower significant bits as you describe would help reduce file size but only by a bit for high ISO images as there would still be large jumps between codes that would conflict with the current Huffman compression tables used. For Canon, one could change the Huffman compression tables as they are embedded in the file (although not all decompression programs respect those), but they are just assumed for most other cameras (at least for Nikon and Pentax).

Modifying the Nikon NEF format would be possible but one would still be stuck with the same assumed Huffman compression tables which would not be optimized for the new less stochastic content.

The advantage of using the DNG format for doing this is that both linearization table and variable Huffman tables embedded in the file are already supported and such support is required for all full specification DNG readers.

The camera identity and metadata is preserved inside the DNG container as long as raw convertors recognize them properly, and such raw convertors have a choice of either using the embedded colour profile or the one they assign to that particular camera model.

That's why I think the DNG container would be the ideal format as a destination format for such a new compression: all one needs to write is a DNG to DNG "lossy" compression converter program.

Regards, GordonBGood

Reply   Reply with quote   Complain
Post (hide subjects)Posted by
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark post MMy threads
Color scheme? Blue / Yellow